Predictive modeling of Han Dynasty beacon towers accounting for spatial heterogeneity: a geographically weighted and GeoShapley-interpretable framework

The beacon-tower network of the Han Dynasty Great Wall played a vital role in safeguarding the Ancient Silk Road, yet many sites are threatened by environmental erosion and human disturbance. To address spatial heterogeneity in archeological landscapes, this study develops a geographically weighted predictive framework for beacon-tower identification in the Shule River Basin. Twelve environmental variables encompassing military defense, logistical support, and communication signaling are constructed to characterize site-selection logic. A geographically weighted random forest combined with Bayesian confidence calibration and spatial conditional random field (GWRF–CRF) captures localized spatial variations, achieving an AUC of 0.887 and outperforming Random Forest, XGBoost, and Logistic Regression. GeoShapley-based interpretation quantifies individual factor contributions and nonlinear interactions, revealing threshold effects for terrain positioning and distance-decay constraints for logistical resources. This framework provides a transferable methodological reference for archeological discovery, risk assessment, and conservation planning in large-scale linear cultural heritage landscapes.